使用MLPRegressor时遭遇KeyError问题求助

我在尝试运行一个MLP回归模型时遇到了KeyError，其他类似案例我都能解决，但这个问题难住我了。我试过调整axis参数和reshape，但都没用。

运行代码后出现如下错误：

Traceback (most recent call last):
File "c:\BAULO\PYTHON\ESTRUCTURAS\P_ML\UNIDAD6\Book16.py", line 14, in <module>
data_y = msu_df[w]
File "C:\Users\Mauri\AppData\Local\Programs\Python\Python310\lib\site-packages\pandas\core\frame.py", line 3810, in __getitem__
indexer = self.columns._get_indexer_strict(key, "columns")[1]
File "C:\Users\Mauri\AppData\Local\Programs\Python\Python310\lib\site-packages\pandas\core\indexes\base.py", line 6111, in _get_indexer_strict
self._raise_if_missing(keyarr, indexer, axis_name)
File "C:\Users\Mauri\AppData\Local\Programs\Python\Python310\lib\site-packages\pandas\core\indexes\base.py", line 6171, in _raise_if_missing
raise KeyError(f"None of [{key}] are in the [{axis_name}]")
KeyError: "None of [Index([('N_Applications',)], dtype='object')] are in the [columns]"

我的代码如下：

import pandas as pd
import numpy as np
from sklearn.neural_network import MLPRegressor

msu_df = pd.read_csv('MSU applications.csv')
msu_df.set_index('Year', drop=True, inplace=True)

X = ['P_Football_Performance','SMAn2']
y = 'N_Applications'

w = np.reshape(y, (1,-1))
data_X = msu_df[X]
data_y = msu_df[w]

mlp = MLPRegressor(hidden_layer_sizes=6, max_iter=100000)
print(mlp.predict(mlp.fit(data_X, data_y)))

问题出在哪？ #

你这错误核心是对目标列名的处理搞复杂啦！原本y就是个普通的字符串列名'N_Applications'，结果你用np.reshape(y, (1,-1))把它转成了一个二维数组。当你拿这个数组去取DataFrame的列时，pandas会把它当成一个带元组的索引来找列，但你的DataFrame里哪有这种奇怪的列名呀，自然就抛出KeyError了。

怎么改？ #

完全不需要给列名做reshape操作，scikit-learn的模型对pandas的Series兼容性很好，直接用原始列名取数就行。修正后的代码如下：

import pandas as pd
import numpy as np
from sklearn.neural_network import MLPRegressor

msu_df = pd.read_csv('MSU applications.csv')
msu_df.set_index('Year', drop=True, inplace=True)

X = ['P_Football_Performance','SMAn2']
y = 'N_Applications'

# 删掉那个多余的reshape步骤！
data_X = msu_df[X]
data_y = msu_df[y]

mlp = MLPRegressor(hidden_layer_sizes=6, max_iter=100000)
# 分开写拟合和预测会更清晰，当然你原来的写法也能跑，但可读性差一点
mlp.fit(data_X, data_y)
print(mlp.predict(data_X))

额外小提示 #

如果之后真的需要把目标变量改成二维数组（比如某些小众模型有要求），那也得先拿到列数据再reshape，比如：

data_y = msu_df[y].values.reshape(-1, 1)

不过对于MLPRegressor来说，一维的Series或者数组都能正常用，所以这步其实没必要做~

备注：内容来源于stack exchange，提问作者Mauricio Luis Vega